SkillAgentSearch skills...

radiology-systematic-review

Design/audit systematic or scoping reviews from protocol through synthesis, certainty and living updates.

Install / Use

npx skills add huang-sir1/radiology-skills --skill radiology-systematic-review

Installs into whichever agent you are using.

About this skill
📄

SKILL.md

Installable skill definition

Quality Score

84/100

Supported Platforms

Universal

Our assessment of radiology-systematic-review

radiology-systematic-review scores 84/100 on our quality scale, 255th of 319 Customer Support skills we index.

Its SKILL.md is 17 KB long, well organised into 9 sections and no code examples: a thorough specification that gives an agent plenty to work with.

With 1,785 GitHub stars, it is one of the more widely adopted skills in the catalogue.

Substance
30/30
Structure
13/20
Description
12/15
Adoption
14/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated about 2 months ago, so radiology-systematic-review is actively maintained.
  • Our last check on 2026-11-05 found the source still online.
  • No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
  • Its trust signals score 88/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Safety scan

No issues found

Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.

AI review by kimi-k2.7-code on 2026-10-01. Automated pattern scan on 2026-10-01. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.

radiology-systematic-review compared with similar skills

All 4 of these similar skills score higher than radiology-systematic-review; compare them before choosing.

SkillScoreStarsUpdatedFormat
radiology-systematic-review (this skill)by huang-sir1841.8k56d agoSKILL.md
taste-skillby Leonxlnx100116.8k8d agoCLAUDE.md
Claude-Code-Game-Studiosby Donchitos10039.7k7d agoCLAUDE.md
x402by browser-use100121.0k3d agoSKILL.md
algorithmic-artby anthropics100177.9k55d agoSKILL.md

Frequently asked questions

How do I install radiology-systematic-review?
Run npx skills add huang-sir1/radiology-skills --skill radiology-systematic-review. The install tabs above show the steps for each supported agent.
Which AI agents does radiology-systematic-review work with?
It is written for Universal, as a SKILL.md file. Other agents that read the same format can often use it too.
Is radiology-systematic-review safe to use?
Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It declares no license and scores 88/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is radiology-systematic-review still maintained?
The repository was last updated about 2 months ago, so radiology-systematic-review is actively maintained.

name: radiology-systematic-review description: "Design/audit systematic or scoping reviews from protocol through synthesis, certainty and living updates."

Systematic review and evidence synthesis

Use this skill to turn a review question into an auditable evidence-synthesis project, or to test whether an existing review can support its conclusion. It is an independent scientific entry point and a cross-cutting mentor/reviewer module. It does not require a manuscript to exist first.

Core stance

  • Protocol first. Freeze the question, eligibility, outcomes, synthesis route, independent unit, subgroup/sensitivity families and deviation policy before result-aware decisions. If work began first, record the protocol as RETROSPECTIVE rather than implying prospective registration.
  • Design before pooling. A forest plot is not the objective. Judge clinical, methodological and statistical compatibility before quantitative synthesis; use a structured synthesis without meta-analysis when pooling would erase the question.
  • One evidence row is not always one independent study. Detect overlapping cohorts, multiple publications, arms, readers, lesions, images, blocks, sections, cells and spots. Preserve the patient/donor/study hierarchy and never inflate precision through pseudoreplication.
  • Risk of bias and applicability are not journal prestige scores. Match the assessment tool to the included design and evaluate domains with evidence anchors. Do not average a fatal domain into a reassuring total score.
  • Constructive review. For every problem, state the criterion, consequence, minimum repair, stronger option and cost, closure evidence, and claim that survives if it remains unresolved.
  • No invention. Never fabricate a searched database, search date, registration, duplicate screening, included study, extracted value, risk-of-bias judgment, effect size, heterogeneity, certainty rating or review result.

Choose the entry mode

| Mode | Use when | Required emphasis | |---|---|---| | protocol | the review is being designed | question framework, eligibility, outcomes, search brief, duplicate processes, pooling gate and prospective decision rules | | audit | a protocol, search log, screening file, extraction sheet, analysis or draft exists | artifact-bound verification, deviations, overlap, missing fields, bias/applicability, reproducibility and atomic repairs | | mentor | a learner needs choices explained while building the review | options, tradeoffs, examples, next decision, minimum viable route and stronger route without silently doing unverified work | | synthesis-handoff | extraction is sufficiently mature for synthesis or meta-analysis | compatible effect rows, hierarchy, covariance/dependence, model candidates, heterogeneity and sensitivity specification for radiology-stats | | writing-handoff | synthesis findings are fixed and need a manuscript package | PRISMA flow inputs, Methods/Results/display jobs, certainty, limitations and claim ceiling for reporting/writing modules |

Also declare review_route from the routing table below and the evidence state of supplied artifacts: PLANNED, AUTHOR_REPORTED, PARTLY_VERIFIED, VERIFIED, or NOT_ASSESSABLE.

Required route selection

Open the evidence-synthesis router and select one primary route. Hybrid reviews may activate more than one route, but keep separate eligibility, extraction and synthesis branches when their estimands differ.

| Route | Typical question | |---|---| | scoping | What concepts, methods, evidence types and gaps exist? | | systematic-narrative | What does comparable evidence show when quantitative pooling is unsuitable? | | diagnostic-accuracy-meta | How accurately does an index test identify the target condition against a reference standard? | | prevalence-incidence-meta | What proportion or rate has a defined state or event in a specified population, setting and period? | | observational-association-meta | What is the bounded association between an exposure/feature and an outcome after preserving design, contrast and adjustment? | | reliability-agreement-method-comparison-meta | How reproducible, concordant or quantitatively interchangeable are readers, measurements, segmentations, devices or methods? | | prediction-radiomics-ai-meta | How well is an imaging/radiomics/AI model developed, validated, calibrated or clinically evaluated? | | prognostic-meta | What is the association between a factor/model and a time-bound outcome? | | intervention-meta | What is the effect of an imaging, AI-guided or biological intervention? | | omics-mechanism-synthesis | What bulk/scRNA/spatial/pathology/perturbation evidence supports a bounded biological explanation? |

Minimal intake

Record available materials and mark missing decision-bearing fields AUTHOR_INPUT_NEEDED:

  1. review objective, intended audience/use, scope and preliminary claim;
  2. route and question framework: PICO, PIRD, PECO/PICOS, CHARMS-like prediction domains, or a concept-context-population framework for scoping reviews;
  3. protocol version/date, registration platform/identifier/status and amendments;
  4. databases/platforms, coverage dates, complete strategies, last-search date and grey-literature policy—reported as a search brief until verified by radiology-search; search-strategy peer-review receipt/status and lifecycle state STATIC, LIVING or RETIRED;
  5. eligibility, outcomes/estimands, time horizons, effect measure, report/study unit and language/ publication-status restrictions;
  6. screening, conflict resolution, exclusion reasons, PRISMA counts and any automation/priority screening contract: tool, model, prompt and version; pilot/calibration evidence; frozen threshold; low-priority exclusion audit and recall safeguard; human override; and final human responsibility;
  7. extraction form, reviewer structure, linked reports/overlapping cohorts and usable effect rows;
  8. risk-of-bias/applicability tool/version, judgments and evidence anchors;
  9. synthesis outputs/code, independent unit, dependence structure, model, heterogeneity, subgroup/sensitivity and certainty evidence.

Required workflow

  1. Classify and frame. Choose the route and define the estimand/question before choosing a synthesis method. Do not force every review into PICO; use PIRD for diagnostic accuracy and prediction-specific domains when model development/validation is central.
  2. Freeze or reconstruct the protocol. Use the protocol template. Record prospective, retrospective, amended or unregistered status and distinguish dated decisions from later deviations.
  3. Create the search handoff. Specify concepts, synonyms, controlled vocabulary needs, eligible designs, databases/platforms, dates and update trigger. radiology-search owns actual database execution, exact query logs, deduplication and corpus acquisition; radiology-citation owns identifier/reference verification. Record whether the complete source-specific strategy was independently peer reviewed, by whom/which role, against which stated criterion, the strategy version reviewed, comments and resolution locators, unresolved issues and whether material amendments require re-review. NOT_REVIEWED is an honest state, not a reason to invent a receipt. Also declare STATIC, LIVING or RETIRED, with update trigger/cadence or retirement reason; a living label does not prove the search is current.
  4. Verify selection integrity. Reconcile retrieved, deduplicated, screened, full-text and included counts; document independent screening, conflict resolution and exclusion reasons. For AI-assisted or priority screening, freeze the exact tool/model/prompt/version and threshold before production use; record pilot/calibration corpus and results; preserve every score, recommendation, exclusion and human override; audit a prespecified sample of low-priority exclusions against a recall safeguard; and name the human who retains final inclusion/exclusion responsibility. If the low-priority audit breaches the frozen safeguard, STOP exclusions, expand human screening and recalibrate without deleting the failed pilot. Preserve a study-family identifier linking multiple reports from one cohort.
  5. Extract for the question and design. Use the extraction matrix. Separate study/report metadata, population, index/exposure/intervention, comparator/reference standard, outcomes, hierarchy, effect estimates and adjustment from reviewer interpretations.
  6. Assess risk of bias and applicability. Open the methods and quality-gates reference, choose a design-matched tool, and use the risk matrix. Preserve domain judgments and supporting anchors; do not invent a numeric quality score.
  7. Run the pooling-feasibility gate before meta-analysis. Test compatibility of population, design, target condition/endpoint, time horizon, index/model/intervention/exposure, comparator/ reference standard, effect definition, threshold, unit and dependence. Return POOL, STRATIFY, NARRATIVE, or STOP_FOR_REPAIR with reasons.
  8. Specify synthesis. For compatible rows, define effect-size transformation, within-study uncertainty/dependence, random/fixed or hierarchical model candidate, heterogeneity, small-study assessment, prespecified subgroup/meta-regression and sensitivity analyses. Hand computation and statistical model implementation to radiology-stats; never claim an analysis ran from a plan.
  9. Handle non-pooling honestly. Group evidence by question-bearing dimensions, tabulate effect direction/magnitude/precision and risk of bias, explain heterogeneity, and state why quantitative synthesis was not justified. Do not vote-count only by statistical significance.
  10. Judge certainty and claim ceiling. Apply GRADE or another verified, question-appropriate certainty framework when relevant, or an explicit confidence rationale when it is not. Do not label an informal impression as GRADE. Separate association, prediction, diagnostic performance, intervention effect, reproducibility and mechanism. Cross-sectional or inferred omics concordance does not establish causality.
  11. Package reporting and writing handoffs. Send current checklist/version audit and PRISMA/ PRISMA-DTA placement to radiology-reporting; send only fixed, traceable evidence and bounded claims to radiology-writing.
  12. Operate a living route when declared. For LIVING, freeze update triggers/cadence, search version, new-record triage, study-family reconciliation, analysis/checklist changes, public version links and retirement criteria. After publication, register corrections, retractions and evidence that can invalidate eligibility, risk-of-bias, synthesis or certainty through the pipeline post-publication lifecycle. LIVING is an operating commitment, not a freshness badge.

Scientific red lines

  • Do not change eligibility, outcomes, thresholds, time horizons, subgroup families or pooling decisions because one option yields a favourable conclusion; record justified amendments.
  • Do not treat reports from the same cohort, multiple thresholds, time points, outcomes, readers, lesions, sections, cells or spots as statistically independent without modelling or selecting the dependence.
  • Do not pool adjusted and unadjusted estimates, development and external-validation performance, per-lesion and per-patient accuracy, incompatible time horizons, or incompatible effect measures merely because they share a label.
  • For

Truncated for display — read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars1.8k
CategoryCustomer
Updated1mo ago
Forks17

Languages

Python

Trust signals

88/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 medium